Overview

I built a fully functional limit order book (LOB) reconstruction engine — the core data structure behind every exchange on Earth. This wasn’t just textbook implementation; I had to handle the edge cases and optimisations that make real market systems work.

What I Built

A production-grade order book that processes add, reduce, and cross messages in real-time while maintaining price-time precedence. The system accurately reflects market state at any point in time, handling the complexity that traders and exchanges deal with every microsecond.

The Hard Parts

1. Crossing Order Logic — The Tricky Case

Most order book implementations assume incoming orders either add to the book or completely execute. I had to handle orders that:

• Execute against multiple existing orders at different price levels
• Partially fill, leaving a residual order in the book
• Must match greedily at best prices first (FIFO at same price)

This meant building a smart matching algorithm that prioritises orders correctly and leaves no ambiguity.

2. Price-Time Precedence — Easy to Get Wrong

Orders at the same price must execute FIFO by arrival time, not by order ID string length. I implemented custom sorting that:

• Groups by price (best prices execute first)
• Breaks ties by arrival time (older orders execute first)
• Preserves distinct order IDs for partial fills

The sorting had to be efficient — not just correct.

3. Partial Fills & Cancellations

When reducing an order by amount X, if X exceeds the existing order, the entire order cancels. If X is less, the order shrinks. I had to:

• Track which orders can be partially reduced
• Maintain order ID uniqueness for lookups
• Never lose state during multi-step operations

Technical Implementation

Data Structure: The order book is a 2D grid where each cell contains order information (price, size, ID). Both bid and ask sides maintained separately for O(1) lookups.

Matching Algorithm:

• Sort by price first (ascending for asks, descending for bids)
• Within same price, maintain FIFO by arrival order
• Match incoming order against best prices first
• Track residual volume for partial fills

Edge Cases Handled:

• Empty book sides (returns None for undefined metrics)
• Orders crossing spread mid-execution
• Reduce messages exceeding order size
• Same-price order priority

Results

Passed all 9 test cases across increasingly complex scenarios:

• ✅ Non-crossing add messages
• ✅ Reduce messages (partial and full)
• ✅ Complex crossing scenarios with multiple levels hit
• ✅ All edge cases handled correctly

Every sample output matched expected results exactly — no discrepancies, no edge case failures.

Why This Matters

Limit order books are fundamental to:

• High-frequency trading — nanosecond-level order management
• Market simulation — backtesting depends on accurate order dynamics
• Exchange infrastructure — matching engines rely on this logic

Getting this right means understanding the entire ecosystem that makes financial markets work.

Tech Stack

Python · Pandas · Data Structures · Financial Engineering

---

COMP226 coursework; foundation for understanding market microstructure and how real exchanges handle orders.